Policy generation method and device based on data fusion, equipment and storage medium

Abstract

The invention relates to the technical field of intelligent decision making, and discloses a strategy generation method, device and equipment based on data fusion and a storage medium, the method comprises the following steps: determining a current user intention through a large language model and a user information query instruction so as to determine a target professional agent, and determining target information through the target professional agent; and generating a target strategy through the target professional agent and the target information. Through the above mode, the application focuses on a specific business scene by constructing the preset professional agent. And understanding a user query instruction through the large language model, and routing the user query instruction to the target professional agent. The target professional agent and the multi-source heterogeneous information base are deeply integrated, heterogeneous information analysis and integration are accurately completed, and a target strategy matched with the user intention is generated. The method can be applied to the business fields of financial science and technology, medical health, old-age care and the like, and the accuracy of the intelligent customer service system for generating the personalized strategy according to the user intention is improved.

Description

Technical Field

[0001] This application relates to the field of intelligent decision-making technology, and in particular to a strategy generation method, apparatus, device and storage medium based on data fusion. Background Technology

[0002] With the deepening development of fintech, financial institutions need to provide comprehensive service solutions to high-net-worth and corporate clients, encompassing asset allocation, market risk warnings, and interpretation of compliance policies. Under the current model, financial advisors or account managers must manually integrate multi-dimensional information such as client transaction records, asset holdings, credit records, and external market sentiment, which is time-consuming and highly dependent on personal experience. Although some financial institutions have established data platforms, problems such as information fragmentation, complex dimensions, and high professional barriers exist, making it difficult to quickly respond to clients' customized needs.

[0003] In the fields of healthcare and elderly care, proactive health management services for the elderly and patients with chronic diseases are becoming increasingly important. Currently, health managers or elderly care consultants need to develop personalized intervention plans based on patients' electronic medical records, physical examination reports, wearable device monitoring data, medical insurance settlement records, and elderly care policies and regulations. This process faces challenges such as severe data silos, difficulties in cross-institutional information collaboration, and rapid updates in medical knowledge, resulting in low service efficiency, homogenized plans, and an inability to meet the diverse and precise health and elderly care needs of each individual.

[0004] Therefore, in the fields of fintech, healthcare, and elderly care, improving the accuracy of intelligent customer service systems in generating personalized strategies based on user intent has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a strategy generation method, apparatus, device, and storage medium based on data fusion to improve the accuracy of intelligent customer service systems in generating personalized strategies according to user intent.

[0006] Firstly, this application provides a strategy generation method based on data fusion, the method comprising: Based on the types of heterogeneous information in the target user's pre-defined multi-source heterogeneous information database, construct at least one pre-defined professional intelligent agent; Receive user information query instructions and determine the current user intent using a pre-trained large language model and the user information query instructions; Based on the current user intent, a target professional intelligent agent is determined from each of the preset professional intelligent agents, and target information matching the current user intent is determined through the target professional intelligent agent. Using the target professional intelligent agent and the target information, a target strategy that matches the current user intent is generated.

[0007] Secondly, this application also provides a strategy generation apparatus based on data fusion, the apparatus comprising: The professional intelligent agent construction module is used to construct at least one preset professional intelligent agent based on the types of heterogeneous information in the preset multi-source heterogeneous information database of the target user. The current user intent determination module is used to receive user information query instructions and determine the current user intent through a pre-trained large language model and the user information query instructions. The target information determination module is used to determine a target professional intelligent agent from each of the preset professional intelligent agents according to the current user intent, and to determine target information that matches the current user intent through the target professional intelligent agent. The target strategy generation module is used to generate a target strategy that matches the current user intent using the target professional intelligent agent and the target information.

[0008] Thirdly, this application also provides a computer device, the computer device including a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the strategy generation method based on data fusion as described above.

[0009] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the data fusion-based strategy generation method described above.

[0010] This application discloses a strategy generation method, apparatus, device, and storage medium based on data fusion. The strategy generation method based on data fusion includes: constructing at least one preset professional intelligent agent based on the types of heterogeneous information in a preset multi-source heterogeneous information database of a target user; receiving a user information query instruction and determining the current user intent through a pre-trained large language model and the user information query instruction; determining a target professional intelligent agent from the preset professional intelligent agents based on the current user intent, and determining target information matching the current user intent through the target professional intelligent agent; and generating a target strategy matching the current user intent through the target professional intelligent agent and the target information. Through the above method, this application focuses on specific types of information and business scenarios by constructing multiple preset professional intelligent agents. It uses a large language model for intent recognition, understands user query instructions, and automatically routes them to the target professional intelligent agent. The target professional intelligent agent is deeply integrated with the multi-source heterogeneous information database, quickly and accurately completing the analysis and integration of heterogeneous information to generate a target strategy matching the user intent. In business fields such as fintech and healthcare / elderly care, this improves the accuracy of intelligent customer service systems in generating personalized strategies based on user intent. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic flowchart of a strategy generation method based on data fusion provided in an embodiment of this application; Figure 2 A schematic block diagram of a data fusion-based strategy generation apparatus provided for embodiments of this application; Figure 3 A schematic block diagram of the structure of a computer device provided for an embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the described order. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0015] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0016] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0017] This application provides a data fusion-based strategy generation method, apparatus, device, and storage medium. The data fusion-based strategy generation method can be applied to intelligent customer service systems by constructing multiple pre-defined specialized intelligent agents, each focusing on specific types of information and business scenarios. It uses a large language model for intent recognition, understands user query commands, and automatically routes them to the target specialized intelligent agent. The target specialized intelligent agent is deeply integrated with a multi-source heterogeneous information database, rapidly and accurately completing the analysis and integration of heterogeneous information to generate target strategies that match user intent. In business areas such as fintech and healthcare / elderly care, this improves the accuracy of intelligent customer service systems in generating personalized strategies based on user intent.

[0018] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0019] Please see Figure 1 , Figure 1 This is a schematic flowchart illustrating a data fusion-based strategy generation method provided in an embodiment of this application. This data fusion-based strategy generation method can be applied to intelligent customer service systems to improve the accuracy of generating personalized strategies based on user intent.

[0020] like Figure 1 As shown, the strategy generation method based on data fusion specifically includes steps S10 to S40.

[0021] Step S10: Construct at least one preset professional intelligent agent based on the types of heterogeneous information in the target user's preset multi-source heterogeneous information database; Specifically, based on the target customer's pre-set multi-source heterogeneous information database, multiple pre-set professional intelligent agents are built for different professional fields or information types, such as company overview analysis intelligent agents, market dynamic monitoring intelligent agents, risk underwriting assessment intelligent agents, etc. Each intelligent agent is specifically responsible for processing and analyzing a specific type of information.

[0022] Step S20: Receive user information query instruction, and determine the current user intent through the pre-trained large language model and the user information query instruction; Specifically, when a user enters an information query command (such as "Please analyze the recent risk situation of XX company and prepare visit materials"), the pre-trained large language model is used to perform deep semantic understanding and parsing of the command, accurately identifying the user's current core intent.

[0023] In one embodiment, taking the fintech business as an example, natural language query commands are received from users (such as account managers, investment advisors, or end customers). For example, an investment advisor might enter the command: "Analyze the investment value of a certain technology company and assess the impact of current market sentiment on it." The pre-trained large language model accurately identified the core entity "a certain technology company". Simultaneously, it extracted the core action instructions "analyze investment value" and "assess the impact of market sentiment". Leveraging its built-in financial knowledge, the large language model understands that "investment value" typically involves financial analysis, valuation models, and growth potential assessment.

[0024] Based on the above analysis, a comprehensive investment analysis report on "a certain technology company" is generated. The analysis dimensions include fundamental analysis (such as financial health and intrinsic valuation) and market analysis (such as public opinion and short-term price drivers), and finally outputs investment recommendations based on multi-dimensional analysis.

[0025] In one embodiment, taking the healthcare and elderly care business as an example, natural language query instructions are received from users (such as elderly people, family members, caregivers, or doctors). For example, a caregiver might input the instruction via voice or text: "The elderly gentleman said he felt dizzy last night, and his blood pressure was a little high this morning. What precautions should be taken for him in this situation? What should be done next?" The pre-trained large language model performs multi-level analysis on the instruction, identifies the core entity "Mr. Wang," and associates it with the individual's health record in the system. Simultaneously, it identifies the key medical information "dizziness" and "high blood pressure" as core symptoms.

[0026] Based on medical knowledge, the pre-trained large language model determines that the core demand of this information query instruction is not a simple factual question and answer, but a request for a comprehensive health risk assessment and personalized intervention plan suggestions. Therefore, "current user intent" can be understood as: "to assess the urgency and risk factors of a specific combination of symptoms (dizziness + hypertension) experienced by a specific user (elderly man), and generate a personalized health management plan that includes monitoring suggestions and action guidance."

[0027] Step S30: Determine the target professional intelligent agent from each of the preset professional intelligent agents according to the current user intent, and determine the target information that matches the current user intent through the target professional intelligent agent; Specifically, the identified "current user intent" is matched with the functional scope of each "preset professional intelligent agent," and the most relevant "target professional intelligent agent" (such as the risk underwriting assessment intelligent agent and the market dynamic monitoring intelligent agent mentioned above) is dispatched. The target professional intelligent agent then retrieves, extracts, and performs preliminary analysis of "target information" that is highly relevant to the current user intent from a multi-source heterogeneous information database in parallel.

[0028] Step S40: Generate a target strategy that matches the current user intent using the target professional intelligent agent and the target information.

[0029] Specifically, the target-oriented intelligent agent fills the target information into its built-in strategy template, uses a large language model to optimize the content and enhance the logic, and generates a complete and executable target strategy (such as generating a corporate risk hedging plan report that includes asset allocation ratios and hedging tools, or a personalized health and wellness plan that includes rehabilitation training programs and nutritional advice), and outputs the strategy in a standardized format for users to use.

[0030] In the fintech business, a multi-source heterogeneous information database was built for a specific user, including their annual reports (financial data), historical insurance policies and claims records (internal data), and industry research reports (external data). Based on these information types, several pre-defined specialized intelligent agents were constructed: Financial risk analysis AI: Specializes in analyzing financial statements, debt ratios, cash flow, and other indicators; Underwriting and claims analysis AI: Familiar with the company's policy terms and skilled at analyzing loss ratios, loss patterns, and historical claims reasons; Market Dynamics Analysis Agent: Focused on tracking industry news, policies and regulations, and market competition information.

[0031] When the user's information query instruction is "Analyze the risk status of a company and generate a visit strategy", the large language model is used to parse the user's information query instruction and identify the following core intents: Core task: Generate a customer visit strategy; Focus area: The client's overall risk profile in the near term, particularly the property insurance risks associated with the new factory; Action requirements: Conduct analysis and provide actionable strategies; Solution Integration Agent: Responsible for integrating the analysis results from various disciplines into a coherent narrative and recommendations.

[0032] Based on the above intent, immediately schedule the three target-specific intelligent agents to work collaboratively: The financial risk analysis agent retrieved the target information from the database: "A company recently issued bonds to expand its new factory, causing its debt-to-equity ratio to rise from 50% to 65%." The insurance claims analysis AI retrieved and analyzed internal data and discovered the target information: "The company has two fire claims records in the old factory area in the past three years, and the area where the new factory is located has a high flood risk coefficient." The market dynamics analysis agent crawls and extracts target information from the internet: "Industry reports indicate that a major competitor of a certain company experienced a major safety incident last month, resulting in a complete production halt."

[0033] Each target-specific intelligent agent submits key insights to the solution integration intelligent agent, which integrates all information based on preset business document templates and the text generation capabilities of a large language model to generate a structured target strategy.

[0034] This embodiment discloses a strategy generation method based on data fusion. The method includes: constructing at least one preset professional intelligent agent based on the types of heterogeneous information in a preset multi-source heterogeneous information database for a target user; receiving a user information query instruction and determining the current user intent using a pre-trained large language model and the query instruction; determining a target professional intelligent agent from the preset professional intelligent agents based on the current user intent, and determining target information matching the current user intent using the target professional intelligent agent; and generating a target strategy matching the current user intent using the target professional intelligent agent and the target information. Through this method, this application focuses on specific types of information and business scenarios by constructing multiple preset professional intelligent agents. It uses a large language model for intent recognition, understands user query instructions, and automatically routes them to the target professional intelligent agent. The target professional intelligent agent is deeply integrated with the multi-source heterogeneous information database, quickly and accurately completing the analysis and integration of heterogeneous information to generate a target strategy matching the user intent. In business areas such as fintech and healthcare / elderly care, this improves the accuracy of intelligent customer service systems in generating personalized strategies based on user intent.

[0035] based on Figure 1 In the illustrated embodiment, step S20 includes: The user information query command is subjected to noise filtering, sensitive word desensitization and referential resolution to generate query command text information; The query command text information is segmented into words to generate at least one basic keyword; The pre-trained large language model is used to classify the intent of each of the basic keywords to generate at least one intent keyword. The current user intent is generated based on the intent keywords.

[0036] Specifically, after receiving the original user information query instruction, it is preprocessed to improve the quality of the instruction: noise filtering removes meaningless characters, spelling errors and colloquial redundancy from the instruction; sensitive word desensitization automatically identifies and blocks or replaces sensitive information involving personal privacy, trade secrets, etc. (such as customer names, specific amounts); and pronoun resolution parses and clarifies the specific objects referred to by pronouns in the instruction (for example, clarifying "the company's latest risks" as "the latest risks of a specific company"), generating query instruction text information.

[0037] The text information is segmented using a precise word segmentation tool, breaking it down into semantically independent lexical units to generate a basic keyword set (e.g., ["analysis", "company", "risk"] from "analyze the risk of a company"). These basic keywords are then used as prompts and input into a pre-trained large language model. Leveraging its powerful semantic understanding capabilities, the model performs in-depth analysis and categorization of the keywords, mapping them to predefined intent categories and outputting more generalized and directional intent keywords (e.g., "property insurance risk assessment", "customer visit plan generation"). Finally, all intent keywords are combined to generate a structured current user intent.

[0038] In a specific embodiment, generating the current user intent based on each of the intent keywords includes: The confidence score of each intent keyword is calculated using the Softmax function. Intent keywords whose confidence level is greater than a preset confidence threshold are identified as target keywords; The pre-trained large language model associates the target keywords to generate the current user intent.

[0039] Specifically, after the pre-trained large language model classifies basic keywords and generates multiple intent keywords (such as "financial analysis", "market competition analysis", and "risk assessment"), the Softmax function is used to calculate the intent keyword confidence score for each intent keyword. The Softmax function transforms the model's original output scores into a probability distribution such that the sum of the confidence scores of all intent keywords is 1. For example, after calculation, the confidence score for "risk assessment" is 0.65, for "financial analysis" it is 0.25, and for "market competition analysis" it is 0.10.

[0040] Keyword filtering is performed by comparing the confidence score of each intent keyword with a preset confidence threshold (e.g., set to 0.2). For example, intent keywords with a confidence score greater than this threshold are identified as target keywords. In this example, "risk assessment" (0.65), "financial analysis" (0.25), and "market competition analysis" (0.10) all exceed the threshold and are therefore retained as target keywords.

[0041] The pre-trained large language model deeply understands the inherent logical relationship between target keywords and ultimately generates the current user intent.

[0042] based on Figure 1 In the illustrated embodiment, step S30 includes: Obtain a preset intent-agent mapping table, and determine at least one professional agent to be determined based on the preset intent-agent mapping table and the current user intent; If the number of the professional intelligent agents to be determined is one, then the professional intelligent agent to be determined is determined as the target professional intelligent agent; If the number of the professional intelligent agents to be determined is at least two, then the target professional intelligent agent is determined from each of the professional intelligent agents to be determined according to the load balancing algorithm; The target information is determined based on the prompt word template of the target professional intelligent agent and the current user intent.

[0043] Specifically, a pre-defined intent-agent mapping table is obtained. This table predefines the correspondence between different business intents and specialized agents (for example, the mapping table specifies that the intent "financial analysis" corresponds to the "financial analysis agent," and the intent "risk assessment" corresponds to the "risk judgment agent"). The current user intent (i.e., "to conduct a comprehensive risk assessment based on the target company's financial data and the market performance of its competitors") is matched with the mapping table. It is found that this intent requires the "financial analysis agent" to provide financial data and the "risk judgment agent" to perform risk calculations. Therefore, two specialized agents to be determined are identified.

[0044] Since there are two agents to be determined, a load balancing algorithm (such as based on the current CPU load or task queue length of each agent service instance) is activated to select the instance with the lighter load as the target professional agent for this execution. The target information is determined based on the prompt template of the target professional agent and the current user intent. For example, if the selected agent is a "risk assessment agent," its prompt template might be "Please assess the risk of [risk type] based on [data dimension A] and [data dimension B] of [company name]." The specific parameters from the intent (company name = "a certain company," data dimension A = "financial data," data dimension B = "competitor market performance," risk type = "comprehensive") are filled into the template to generate the target information.

[0045] based on Figure 1 In the illustrated embodiment, step S40 includes: The target professional intelligent agent performs contextual understanding processing on the target information to generate a preliminary strategy associated with the current user intent; A target strategy generation template is determined based on the target professional intelligent agent; The target strategy is generated by performing a deep analysis of the initial strategy using the target strategy.

[0046] Specifically, the target-specific intelligent agent performs contextual understanding processing on the target information, analyzes the causal relationships and business logic between data, and generates a preliminary strategy. Based on the type of the target-specific intelligent agent, a target strategy generation template is determined, such as a standard "Corporate Credit Risk Assessment Report Template." This template has a predefined structure, including "Risk Level," "Major Risk Points," "Credit Granting Recommendations," and "Post-Loan Management Focus." The preliminary strategy is deeply analyzed using the target strategy generation template, and the key information from the preliminary strategy is filled in and refined into the corresponding parts of the template to generate the target strategy.

[0047] based on Figure 1 In the illustrated embodiment, step S10 includes: Based on the type of each heterogeneous information, a preset business scenario matching each heterogeneous information is determined; The current task objective is determined based on the preset business scenario, and the task execution process is determined based on the current task objective; Based on the task execution process, construct each of the preset professional intelligent agents.

[0048] Specifically, based on the inherent characteristics and business value of various types of information in multi-source heterogeneous information databases, pre-defined business scenarios are determined to match them. For example, time-series financial data (such as financial statements) is determined to be suitable for the "corporate financial health analysis and prediction" scenario; unstructured public opinion text data (such as news and research reports) is determined to be suitable for the "market risk and opportunity insight" scenario; and graph-based relational data (such as equity structure and supply chain information) is determined to be suitable for the "complex risk transmission analysis" scenario.

[0049] For each defined business scenario, its "current task objective" is clarified, and a specific "task execution process" is designed accordingly. Taking the "complex risk transmission analysis" scenario as an example, its current task objective is set as "identifying and quantifying the potential collateral risks that the target company may incur due to related parties," and the resulting task execution process is as follows: 1) Identify the relationship between the core enterprise and its affiliates; 2) Monitor risk events involving related parties; 3) Simulate the risk transmission path and the degree of impact.

[0050] The corresponding "risk transmission analysis intelligent agent" will be constructed strictly in accordance with this task execution process.

[0051] based on Figure 1 In the illustrated embodiment, the steps preceding step S10 include: The target user's basic information, user risk information, and user business information are collected through an information retrieval model. The user basic information, the user risk information, and the user business information are vectorized and encoded to generate feature vectors for basic information, risk information, and business information, respectively. A knowledge graph is generated based on the basic information feature vector, the risk information feature vector, and the business information feature vector, and the preset multi-source heterogeneous information database is constructed based on the knowledge graph.

[0052] Specifically, information retrieval models are used to collect basic user information (such as company name, size, and legal representative), user risk information (such as financial indicators, legal proceedings, and administrative penalties), and user business information (such as main business, supply chain relationships, and market share) from various internal and external data sources. These three types of heterogeneous information are then vectorized and encoded. For example, different embedding models are used to convert unstructured text, tables, and other information into numerical vector representations, generating corresponding feature vectors for basic information, risk information, and business information, respectively.

[0053] Using the target user as the core entity, these feature vectors are used as the features of the nodes. Entity recognition and relation extraction technologies are used to construct a knowledge graph, and a pre-defined multi-source heterogeneous information database is built based on this knowledge graph.

[0054] Please see Figure 2 , Figure 2 This application provides a schematic block diagram of a data fusion-based strategy generation apparatus, which is used to execute the aforementioned data fusion-based strategy generation method. The data fusion-based strategy generation apparatus can be configured on a server.

[0055] like Figure 2 As shown, the data fusion-based strategy generation device 400 includes: The professional intelligent agent construction module 410 is used to construct at least one preset professional intelligent agent based on the types of heterogeneous information in the preset multi-source heterogeneous information database of the target user. The current user intent determination module 420 is used to receive user information query instructions and determine the current user intent through a pre-trained large language model and the user information query instructions. The target information determination module 430 is used to determine a target professional intelligent agent from each of the preset professional intelligent agents according to the current user intent, and to determine target information that matches the current user intent through the target professional intelligent agent. The target strategy generation module 440 is used to generate a target strategy that matches the current user intent using the target professional intelligent agent and the target information.

[0056] Furthermore, the current user intent determination module 420 includes: The query instruction text information generation unit is used to perform noise filtering, sensitive word desensitization and referential resolution on the user information query instruction to generate query instruction text information. The basic keyword generation unit is used to perform word segmentation on the query command text information to generate at least one basic keyword. The intent keyword generation unit is used to classify the intent of each of the basic keywords using the pre-trained large language model, and generate at least one intent keyword. The current user intent generation unit is used to generate the current user intent based on each of the intent keywords.

[0057] Furthermore, the current user intent generation unit includes: The intent keyword confidence calculation subunit is used to calculate the intent keyword confidence of each intent keyword using the Softmax function; The target keyword determination subunit is used to determine the intent keywords whose confidence level is greater than a preset confidence threshold as target keywords; The current user intent generation subunit is used to associate the target keywords through the pre-trained large language model to generate the current user intent.

[0058] Furthermore, the target information determination module 430 includes: The undetermined professional intelligent agent determination unit is used to obtain a preset intent-intent mapping table, and determine at least one undetermined professional intelligent agent through the preset intent-intent mapping table and the current user intent; A target professional intelligent agent determination unit is used to determine the undetermined professional intelligent agent as the target professional intelligent agent if the number of undetermined professional intelligent agents is one. A target professional intelligent agent determination unit is used to determine the target professional intelligent agent from the undetermined professional intelligent agents according to a load balancing algorithm if the number of undetermined professional intelligent agents is at least two. The target information determination unit is used to determine the target information based on the prompt word template of the target professional intelligent agent and the current user intent.

[0059] Furthermore, the target strategy generation module 440 includes: The preliminary strategy generation unit is used to perform contextual understanding processing on the target information through the target professional intelligent agent to generate a preliminary strategy associated with the current user intent; The target strategy generation template determination unit is used to determine the target strategy generation template based on the target professional intelligent agent. The target strategy generation unit is used to perform in-depth analysis of the preliminary strategy through the target strategy generation target to generate the target strategy.

[0060] Furthermore, the specialized intelligent agent construction module 410 includes: A preset business scenario determination unit is used to determine a preset business scenario that matches each of the heterogeneous information based on the type of each heterogeneous information. The task execution process determination unit is used to determine the current task objective based on the preset business scenario, and to determine the task execution process based on the current task objective; A preset professional intelligent agent construction unit is used to construct each preset professional intelligent agent according to the task execution process.

[0061] Furthermore, the data fusion-based strategy generation device 400 includes: The heterogeneous information acquisition module is used to collect the target user's basic information, user risk information, and user business information through an information retrieval model. The feature vector conversion module is used to perform vectorized encoding processing on the user basic information, the user risk information and the user business information to generate basic information feature vectors, risk information feature vectors and business information feature vectors, respectively. The multi-source heterogeneous information database construction module is used to generate a knowledge graph based on the basic information feature vector, the risk information feature vector, and the business information feature vector, and to construct the preset multi-source heterogeneous information database based on the knowledge graph.

[0062] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the above-described apparatus and modules can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0063] The aforementioned device can be implemented as a computer program, which can be used in, for example... Figure 3 It runs on the computer device shown.

[0064] Please see Figure 3 , Figure 3 This is a schematic block diagram illustrating the structure of a computer device according to an embodiment of this application. The computer device may be a server.

[0065] See Figure 3 The computer device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0066] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any data fusion-based strategy generation method.

[0067] The processor provides computing and control capabilities, supporting the operation of the entire computer device.

[0068] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any data fusion-based strategy generation method.

[0069] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 3The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0070] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0071] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps: Based on the types of heterogeneous information in the target user's pre-defined multi-source heterogeneous information database, construct at least one pre-defined professional intelligent agent; Receive user information query instructions and determine the current user intent using a pre-trained large language model and the user information query instructions; Based on the current user intent, a target professional intelligent agent is determined from each of the preset professional intelligent agents, and target information matching the current user intent is determined through the target professional intelligent agent. Using the target professional intelligent agent and the target information, a target strategy that matches the current user intent is generated.

[0072] In one embodiment, a user information query instruction is received, and the current user intent is determined using a pre-trained large language model and the user information query instruction, for the purpose of: The user information query command is subjected to noise filtering, sensitive word desensitization and referential resolution to generate query command text information; The query command text information is segmented into words to generate at least one basic keyword; The pre-trained large language model is used to classify the intent of each of the basic keywords to generate at least one intent keyword. The current user intent is generated based on the intent keywords.

[0073] In one embodiment, the current user intent is generated based on each of the intent keywords, for the purpose of: The confidence score of each intent keyword is calculated using the Softmax function. Intent keywords whose confidence level is greater than a preset confidence threshold are identified as target keywords; The pre-trained large language model associates the target keywords to generate the current user intent.

[0074] In one embodiment, a target professional intelligent agent is determined from each of the preset professional intelligent agents based on the current user intent, and target information matching the current user intent is determined through the target professional intelligent agent, for the purpose of: Obtain a preset intent-agent mapping table, and determine at least one professional agent to be determined based on the preset intent-agent mapping table and the current user intent; If the number of the professional intelligent agents to be determined is one, then the professional intelligent agent to be determined is determined as the target professional intelligent agent; If the number of the professional intelligent agents to be determined is at least two, then the target professional intelligent agent is determined from each of the professional intelligent agents to be determined according to the load balancing algorithm; The target information is determined based on the prompt word template of the target professional intelligent agent and the current user intent.

[0075] In one embodiment, a target strategy matching the current user intent is generated using the target professional intelligent agent and the target information, for the purpose of: The target professional intelligent agent performs contextual understanding processing on the target information to generate a preliminary strategy associated with the current user intent; A target strategy generation template is determined based on the target professional intelligent agent; The target strategy is generated by performing a deep analysis of the initial strategy using the target strategy.

[0076] In one embodiment, at least one pre-defined specialized intelligent agent is constructed based on the types of heterogeneous information in a pre-defined multi-source heterogeneous information database of the target user, for the purpose of: Based on the type of each heterogeneous information, a preset business scenario matching each heterogeneous information is determined; The current task objective is determined based on the preset business scenario, and the task execution process is determined based on the current task objective; Based on the task execution process, construct each of the preset professional intelligent agents.

[0077] In one embodiment, before constructing at least one preset professional intelligent agent based on the types of heterogeneous information in a preset multi-source heterogeneous information database for the target user, the following is implemented: The target user's basic information, user risk information, and user business information are collected through an information retrieval model. The user basic information, the user risk information, and the user business information are vectorized and encoded to generate feature vectors for basic information, risk information, and business information, respectively. A knowledge graph is generated based on the basic information feature vector, the risk information feature vector, and the business information feature vector, and the preset multi-source heterogeneous information database is constructed based on the knowledge graph.

[0078] The embodiments of this application also provide a computer-readable storage medium storing a computer program, the computer program including program instructions, and the processor executing the program instructions to implement any of the data fusion-based strategy generation methods provided in the embodiments of this application.

[0079] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiments, such as the hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, SmartMedia Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the computer device.

[0080] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A strategy generation method based on data fusion, characterized in that, include: Based on the types of heterogeneous information in the target user's pre-defined multi-source heterogeneous information database, construct at least one pre-defined professional intelligent agent; Receive user information query instructions and determine the current user intent using a pre-trained large language model and the user information query instructions; Based on the current user intent, a target professional intelligent agent is determined from each of the preset professional intelligent agents, and target information matching the current user intent is determined through the target professional intelligent agent. Using the target professional intelligent agent and the target information, a target strategy that matches the current user intent is generated.

2. The strategy generation method based on data fusion according to claim 1, characterized in that, The step of receiving a user information query instruction and determining the current user intent using a pre-trained large language model and the user information query instruction includes: The user information query command is subjected to noise filtering, sensitive word desensitization and referential resolution to generate query command text information; The query command text information is segmented into words to generate at least one basic keyword; The pre-trained large language model is used to classify the intent of each of the basic keywords to generate at least one intent keyword. The current user intent is generated based on the intent keywords.

3. The strategy generation method based on data fusion according to claim 2, characterized in that, The step of generating the current user intent based on each intent keyword includes: The confidence score of each intent keyword is calculated using the Softmax function. Intent keywords whose confidence level is greater than a preset confidence threshold are identified as target keywords; The pre-trained large language model associates the target keywords to generate the current user intent.

4. The strategy generation method based on data fusion according to claim 1, characterized in that, The step of determining a target professional intelligent agent from among the preset professional intelligent agents based on the current user intent, and determining target information matching the current user intent through the target professional intelligent agent, includes: Obtain a preset intent-agent mapping table, and determine at least one professional agent to be determined based on the preset intent-agent mapping table and the current user intent; If the number of the professional intelligent agents to be determined is one, then the professional intelligent agent to be determined is determined as the target professional intelligent agent; If the number of the professional intelligent agents to be determined is at least two, then the target professional intelligent agent is determined from each of the professional intelligent agents to be determined according to the load balancing algorithm; The target information is determined based on the prompt word template of the target professional intelligent agent and the current user intent.

5. The strategy generation method based on data fusion according to claim 1, characterized in that, The step of generating a target strategy that matches the current user intent using the target professional intelligent agent and the target information includes: The target professional intelligent agent performs contextual understanding processing on the target information to generate a preliminary strategy associated with the current user intent; A target strategy generation template is determined based on the target professional intelligent agent; The target strategy is generated by performing a deep analysis of the initial strategy using the target strategy.

6. The strategy generation method based on data fusion according to claim 1, characterized in that, The step of constructing at least one preset professional intelligent agent based on the types of heterogeneous information in the target user's preset multi-source heterogeneous information database includes: Based on the type of each heterogeneous information, a preset business scenario matching each heterogeneous information is determined; The current task objective is determined based on the preset business scenario, and the task execution process is determined based on the current task objective; Based on the task execution process, construct each of the preset professional intelligent agents.

7. The strategy generation method based on data fusion according to any one of claims 1 to 6, characterized in that, Before constructing at least one preset professional intelligent agent based on the types of heterogeneous information in the target user's preset multi-source heterogeneous information database, the following steps are included: The target user's basic information, user risk information, and user business information are collected through an information retrieval model. The user basic information, the user risk information, and the user business information are vectorized and encoded to generate feature vectors for basic information, risk information, and business information, respectively. A knowledge graph is generated based on the basic information feature vector, the risk information feature vector, and the business information feature vector, and the preset multi-source heterogeneous information database is constructed based on the knowledge graph.

8. A strategy generation device based on data fusion, characterized in that, include: The professional intelligent agent construction module is used to construct at least one preset professional intelligent agent based on the types of heterogeneous information in the preset multi-source heterogeneous information database of the target user. The current user intent determination module is used to receive user information query instructions and determine the current user intent through a pre-trained large language model and the user information query instructions. The target information determination module is used to determine a target professional intelligent agent from each of the preset professional intelligent agents according to the current user intent, and to determine target information that matches the current user intent through the target professional intelligent agent. The target strategy generation module is used to generate a target strategy that matches the current user intent using the target professional intelligent agent and the target information.

9. A computer device, characterized in that, The computer device includes a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program and, in executing the computer program, implement the data fusion-based strategy generation method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to implement the data fusion-based strategy generation method as described in any one of claims 1 to 7.

Images